A Tale of Millis and Nanos: Time Measurements in Virtual and Physical Machines

摘要

Cloud computing makes large infrastructure capacities available to users in a flexible and affordable fashion, which is of specific interest to scientists for conducting experiments. Unfortunately, our past research has provided first indications that virtual machines - the most popular type of cloud-based infrastructure - have substantial deficits with respect to time measurements, which are an important tool for researchers. In this paper, we provide a detailed analysis on the accuracy of time measurements based on various machine configurations. They cover influence factors such as machine type, virtualization solution, and programming language. The results indicate that not the use of virtualization as such, but the potentially uncontrollable utilization of the physical host is a decisive factor for the accuracy of time measurements. Different virtualization solutions and programming languages play an inferior role. Our findings, along with the publicly released tool TiMeAcE.KOM, can provide a valuable decision support for researchers in the selection and configuration of cloud-based experimental infrastructures.